Canon Admissibility Failure Memo (CAFM)
CAFM here means Canon Admissibility Failure Memo (not Computer-Aided Facility Management).
This page publishes an illustrative Canon memo applied to a real-world, regulated decision pipeline. It is provided to show the shape of Canon outputs: the identification of load-bearing abstractions, undeclared hand-offs, and recorded absence.
This exemplar is a non-interpretive structural diagnostic. It does not assess compliance, correctness, quality, or adequacy. It operates on publicly stated obligations and commonly deployed pipeline structures in UK consumer lending. Where the memo notes an undeclared hand-off, it is describing a structural dependency (a reliance on silent reconstruction), not alleging wrongdoing.
This exemplar is anchored to the FCA Consumer Credit sourcebook’s creditworthiness assessment requirements (CONC 5.2A). Canon does not interpret legal meaning. It treats these as publicly stated constraints that shape the authority claims made by regulated lending pipelines.
Primary source: FCA Handbook – CONC 5.2A
Canon’s core finding across multiple domains is consistent: high-stakes artefacts frequently rely on structural silence at the point where responsibility should be explicitly representable. This consumer credit exemplar demonstrates the same structural pattern without asserting any new numeric prevalence claim for lending filings.
It identifies the load-bearing abstraction that converts an AI output into an authoritative credit decision, then records the undeclared interpretive hand-off where responsibility becomes unrepresented unless explicitly assigned.
This exemplar does not assert that any specific lender is non-compliant, nor does it provide recommendations. It only records structural dependencies relevant to accountability claims (e.g., explainability + human oversight).
When regulators and internal governance require firms to evidence outcomes, the weak point is often not data availability but representation: whether accountability can be demonstrated from what is explicitly recorded, without relying on post-hoc narrative assembly. Canon treats that boundary as a structural requirement.
In practice, structurally incomplete files generate downstream workload: manual review, complaint handling, audit reconstruction, and “what did we rely on?” rework under time pressure. Canon’s move is upstream: make stated vs not stated visible early, so governance artefacts are built on explicit evidence boundaries rather than assumptions.
Canon can be deployed as a deterministic validation service upstream of reporting and AI summarisation pipelines. Given an input record (document or data packet), it returns a machine-readable statement of what is explicitly present vs not stated (e.g., actor, action, time, outcome), plus a traceable audit log. It does not infer missing details and does not generate new claims.
Where LLMs are used, Canon functions as a hard evidence boundary: it prevents unsupported fields from entering permanent records by making absence explicit. The output remains deterministic and reproducible.
- FCA Handbook – CONC 5.2A (Creditworthiness assessment)
- FCA Policy Statement PS18/19 – Assessing creditworthiness in consumer credit (PDF)
- FCA Handbook – PRIN 2A.4 (Consumer Duty: price and value outcome)
- ICO – UK GDPR guidance: automated decision-making and profiling
- Legislation.gov.uk – GDPR Article 22 (automated individual decision-making)
These sources are provided for provenance only. Canon does not interpret legal meaning; it treats these as publicly stated constraints that shape the authority claims made by regulated decision pipelines.
For exemplars, the recommended default is: HTML page for discovery + PDF for sharing. The HTML page is indexable, linkable, and accessible; the PDF is stable, printable, and “forwardable” in institutional workflows.
For a lender, Canon establishes a defensible, auditable boundary between what is explicitly evidenced and what would otherwise be silently reconstructed at the point of accountability (e.g., “explainability + human oversight”). This reduces governance fragility under audit, complaint, and supervisory scrutiny by making responsibility hand-offs representable rather than assumed.
Canon does not guarantee regulatory outcomes. It identifies structural incompleteness where authority claims depend on undeclared reconstruction.
(In)Canon identifies structure and reports stated vs not stated. It does not assess meaning, correctness, quality, compliance, or adequacy.